[Diagnostic value of infrared thermography combined with support vector machine in preoperative prediction of cervical lymph node metastasis in oral cancer].

Objective: To evaluate the value of a prediction model based on infrared thermography (IRT) features combined with a support vector machine (SVM) algorithm for the preoperative non-invasive diagnosis of cervical lymph node metastasis in oral squamous cell carcinoma (OSCC), with particular attention given to its potential application in clinically node-negative (cN0) patients. Methods: Clinical data of 150 patients with OSCC admitted to the First Affiliated Hospital of Bengbu Medical University from January 2021 to December 2024 were retrospectively reviewed, including 85 males and 65 females, with a mean age of (59.0±12.6) years. Preoperative cervical IRT data were analyzed, and features such as temperature, texture and vascular morphology were extracted to calculate the differences in features between the affected side and the contralateral healthy side. Analysis of variance (ANOVA) and LASSO regression were used to screen core variables for the construction of a SVM prediction model. All subjects were randomly divided into a training set (105 cases) and an internal test set (45 cases) at a ratio of 7∶3, and propensity score matching (PSM) was performed to balance the baseline characteristics of the training set. Ten-fold cross-validation was used to optimize the model hyperparameters. A total of 30 prospectively recruited OSCC patients were included as the independent validation set. The diagnostic performance of the model was assessed in both the internal test set and independent validation set, and the diagnostic efficacy was further compared with contrast-enhanced computed tomography (CECT). Results: Five variables were ultimately selected for modeling: clinical T stage, primary tumor location in the hard palate, maximum temperature difference (ΔTmax), contrast difference (ΔContrast), and depth of invasion (DOI)>10 mm. In the test set, the SVM model achieved an accuracy of 84.4%, sensitivity of 91.7%, specificity of 76.2%, and an AUC of 0.931. In the prospective validation set, it showed an accuracy of 83.3%, sensitivity of 88.9%, specificity of 75.0%, and an AUC of 0.852. Compared with CECT, there was no statistically significant difference in diagnostic efficacy (χ2=0.115, P>0.05), and the SVM model showed potential as an effective auxiliary tool to enhance the diagnostic performance of CECT. Conclusion: The combination of infrared thermography and an SVM model can effectively predict cervical lymph node metastasis in OSCC, demonstrating high diagnostic efficacy and good clinical applicability. This can serve as a non-invasive auxiliary tool for preoperative assessment of lymph node metastasis status.